Add a method ignore certain wells for training/validation

[edyoshikun]

Given our limited storage on the HPC and the data duplication that happens between the processed data and the training data, it might be helpful to have a variable to ignore certain wells from the HCSDataloader.

For example, for an experiment where we had multiple wells but wanted to use 4 out of 8 for training, we currently create a new store with these wells (data duplication).

It is common on the mantis pipeline to have wells that are used as fiducials, so we should implement add this feature soon.